Two lambdas that serve as a test bench to illustrate memory consumption issues with NR serverless monitoring running on lambdas that generate a lot of spans.
- update the
endpoint
variable in each lambda to point to a webserver that will give back 404s. We found that hitting a simple webserver we spun up w/o rate-limiting enforced with a parameterized route that will give 404 to be simplest way to do this. - Optional: test setup with SAM-Local. Debug configurations for the stress test and control lambdas are supplied, along with run scripts. You will need
sam cli
anddocker
installed. - Deploy lambads to your AWS acct
- We were able to pretty consistently blow out the memory of a 1GB lambda using the following env variables
- RPM = 6000 (essentially allow lambda to run unbounded)
- TIME_DOING_WORK = 850 (make sure you configure lambda for max runtime of 15 minutes)